This week, Google's DeepMind introduced AlphaFold 3, a groundbreaking model that can accurately predict the structure of proteins, DNA, RNA, and their interactions. The previous AlphaFold models were already significant breakthroughs, with the potential for their creators to receive Nobel prizes.

AlphaFold 3 takes this a step further by accurately predicting how these various molecules will interact with each other, and even finding new ways of combining molecules that scientists haven't tried before. This could lead to an explosion of new drug discoveries, new vaccines, and deeper insights into how the human body works.

The impact of this announcement on the medical world and drug discovery is expected to be substantial in the next couple of years, with the potential for AI-designed drugs to enter clinical trials.

Mustafa Suleyman, the co-founder of Anthropic and former DeepMind executive, has joined Microsoft to lead the development of a new large language model called MAI1. This model is rumored to have a staggering 500 billion parameters, making it a potential contender to OpenAI's GPT-4.

Microsoft's move to build its own large language model is a strategic one. The company has historically relied on partnerships with OpenAI and Meta (through their open-source LLAMA models) for its AI capabilities. However, by developing MAI1 in-house, Microsoft is ensuring it has a diversified AI portfolio and is not solely dependent on external providers.

The development of MAI1 also suggests that Microsoft is positioning itself to be less reliant on OpenAI, should the terms of their partnership change in the future. This could be a preemptive measure to maintain Microsoft's AI capabilities and competitiveness, regardless of how its relationship with OpenAI evolves.

With a reported 500 billion parameters, MAI1 would be one of the largest language models publicly available, rivaling the capabilities of GPT-4. This scale could enable Microsoft to push the boundaries of what is possible with large language models, potentially leading to groundbreaking advancements in areas such as natural language processing, generation, and reasoning.

As the AI landscape continues to evolve rapidly, Microsoft's investment in MAI1 demonstrates its commitment to staying at the forefront of the field. This move could have significant implications for the broader AI ecosystem, as it sets the stage for increased competition and innovation in the large language model space.

This week, Stack Overflow and Open AI announced a partnership to strengthen the world's most popular large language models. Since the release of ChatGPT, the traffic to Stack Overflow has been declining as people found the AI tool to be quite capable at coding tasks.

To address this, Stack Overflow tried to launch its own AI model, but it did not gain much traction. Now, the partnership with Open AI aims to improve the coding abilities of models like GPT-4 and future generations.

The key points are:

• Open AI will be able to train on Stack Overflow's vast database of coding data, which should significantly enhance the coding capabilities of their language models.
• However, some users who have contributed answers to Stack Overflow are not happy about this, as the platform's terms grant them the right to retain all answers, essentially allowing Open AI to leverage this content.
• Stack Overflow has started suspending accounts of users who are deleting their answers in response to the partnership announcement.

This collaboration highlights Open AI's strategy to continuously improve their models by leveraging diverse datasets and partnerships. The improved coding abilities of their language models could have significant implications for the future of software development and programming.

Last week, a mysterious chatbot known as GPT-2 popped up inside the Limbs.org chatbot arena, only to be quickly removed. However, this week, the chatbot has returned, and people are eager to explore its capabilities.

The prevailing theory is that this chatbot is some kind of model from OpenAI, the company behind the popular ChatGPT. While OpenAI has been tight-lipped about the chatbot, they have been dropping cryptic memes on Twitter, hinting at their involvement.

This new model, which is still shrouded in mystery, has been able to outperform ChatGPT in certain tasks, leading to speculation about its potential. Some users have reported that the chatbot is able to engage in more natural and coherent conversations, while others have noted its impressive ability to generate creative and insightful responses.

As the AI community continues to explore the capabilities of this mysterious chatbot, it's clear that the field of language models is rapidly evolving. With the potential for new breakthroughs in areas like drug discovery, coding, and even search engines, the future of AI is shaping up to be an exciting and transformative one.

In their recent blog post, Open AI has claimed that they respect the choices of creators and content owners when it comes to AI. To address this, they are building a "media manager" platform that will allow content owners to manage how their works are used in AI.

The goal is to have this tool in place by 2025, which will give content creators the ability to exclude their content from being trained in Open AI's systems. This is a significant move, as it aims to address concerns around the use of copyrighted or sensitive material in AI models.

By providing this level of control, Open AI is attempting to strike a balance between the advancement of AI technology and the rights of content creators. This approach could set a precedent for other AI companies to follow, as the industry grapples with the ethical implications of training models on vast amounts of online data.

Overall, this announcement from Open AI represents a step towards greater transparency and collaboration with the creative community, as the AI landscape continues to evolve.

If you have an online business or are trying to grow an email list, it can be challenging to determine the right words and style to create an effective landing page that encourages people to sign up or make a purchase. That's why I've partnered with HubSpot to showcase their custom GPT-powered landing page creator, which is completely free for all ChatGPT Plus subscribers.

The process is simple. You start by using the pre-written prompt to create a landing page. The tool will then ask you a few questions to ensure it has all the necessary details, such as a brief description of your site or newsletter, what makes it stand out, and the desired call-to-action.

After providing this information, the custom GPT model will generate sales copy for your landing page. You can review the copy and make any desired tweaks before confirming that you're ready to create the page. Within a couple of seconds, you'll have a link to your new, well-designed landing page that you can further customize with relevant images.

This HubSpot tool makes it incredibly easy and fast to set up a landing page that is optimized to grow your mailing list or drive sales. If you have a ChatGPT Plus account, I've included the link in the description so you can try it out for yourself. Alternatively, you can scan the QR code on the screen to access the tool directly.

This week, Gradient AI announced that they managed to get a 4 million token context length out of the Llama 38B model. This is the largest context window of any openly available large language model.

A 4 million token context window is essentially 3 million words of input and output that the large language model can process. To put this into perspective, the entire Harry Potter book series is a little over 1 million words. Theoretically, you could feed in three times the size of the entire Harry Potter series and get responses based on that context window.

The fact that Gradient AI was able to achieve this with the open-source Llama 38B model is quite impressive. They have shared their work on Hugging Face, allowing anyone with the right knowledge to download and run the model with the 4 million token context length.

This development suggests that the concept of a "context window" may become less relevant in the coming years, as the context windows of large language models continue to grow exponentially. The ability to process such vast amounts of contextual information could lead to significant advancements in various applications of these models.

Krea AI has launched a new feature that allows users to create captivating videos by blending and morphing between two images. This innovative technology, powered by open-source tools like Animate Diff, enables users to generate dynamic, frame-interpolated videos from static images.

The examples showcased demonstrate the versatility of this feature. Users can seamlessly transition between fantastical scenes, cities morphing into one another, and even surreal animations with a demonic figure transforming into a house. The smooth frame interpolation and creative blending create a mesmerizing visual experience.

While Krea AI's video feature is currently in a limited access phase, the company plans to open it up to more users in the near future. This technology presents exciting possibilities for applications in areas such as music videos, visual effects, and artistic expression, allowing creators to bring their ideas to life in a dynamic, AI-powered format.

This week, Sean Roston, who works with OpenAI on developer support, shared a video showcasing Sora's ability to change the character in a video without affecting the rest of the scene.

The video demonstrates this by showing the same alley scene with different characters - a man, a woman, an older man, and a robot. While the background and environment remain largely the same, the model is able to seamlessly swap out the character in each frame.

This feature allows users to create videos where they can change the character without having to redo the entire scene. If you make a video you really like, you can simply swap out the character and keep the rest of the scene intact.

The ability to inpaint different characters into the same environment is an impressive capability of Sora, OpenAI's upcoming video generation model. This technology could be useful for a variety of applications, from filmmaking to video game development, where the ability to easily modify characters within a scene can save time and effort.

As we await the public release of Sora, this demonstration provides a glimpse into the model's potential and the advancements being made in AI-powered video generation.

This week, the audio generation tool Audeo, which has been praised for its realistic music output, rolled out a new update. The key feature of this update is the addition of an "audio inpainting" capability.

With this new feature, users can now select a specific portion of a generated song and have Audeo modify just that selected chunk, without affecting the rest of the audio. This allows users to fine-tune and improve parts of a song they are not fully satisfied with, rather than having to regenerate the entire piece.

The audio inpainting feature gives users more control and flexibility in shaping the final output of their Audeo-generated music. It addresses a common pain point where users may like most of a generated song but want to tweak a specific section.

This update demonstrates Anthropic's commitment to continuously improving Audeo and providing users with more advanced audio manipulation capabilities. As AI-generated music continues to advance, tools like Audeo's audio inpainting will become increasingly valuable for musicians and creators looking to leverage these technologies in their workflows.

This week, 11 Labs, a company known for its work in realistic voice cloning, audio translations, and sound effects, appears to be getting into the AI music generation game.

In a recent tweet, 11 Labs shared an early preview of their AI-generated music. The tweet included several examples, where all the songs were generated with a single text prompt, with no further edits.

One of the examples shared sounded remarkably similar to the style of Tom Petty, showcasing the impressive capabilities of 11 Labs' AI music generation technology.

While the technology is still in its early stages, the ability to generate realistic-sounding music with just a text prompt is a significant development in the field of AI-powered audio creation. As 11 Labs continues to refine and expand their offerings, it will be exciting to see how this technology evolves and what new possibilities it unlocks for musicians, producers, and audio creators.

The coming weeks are set to be a whirlwind of AI announcements, with major tech events like Google I/O, Microsoft Build, and Apple's WWDC all on the horizon. While this week may have seemed slower in comparison, the groundbreaking news around AlphaFold 3 and Microsoft's new large language model are clear indicators of the rapid advancements happening in the AI space.

As the industry continues to evolve, we can expect to see a flood of new AI-powered tools, models, and capabilities emerge. From improved protein structure prediction to more advanced language models and AI-generated content, the pace of innovation shows no signs of slowing down.

Staying informed and keeping up with the latest developments will be crucial for anyone interested in or working within the AI field. By subscribing to channels like this one and exploring resources like the Next Wave podcast, you can ensure you're always on the cutting edge of the AI revolution.